부산대학교 도서관

ObjectiveTo explore the correlation between word frequency and 17 items of the Hamilton Depression Scale (HAMD-17) in assessing the severity of depression in clinical interviews.MethodsThis study included 70 patients with major depressive disorder (MDD) who were hospitalized in the Beijing Anding Hospital. Clinicians interviewed eligible patients, collected general information, disease symptoms, duration, and scored patients with HAMD-17. The words used by the patients during the interview were classified and extracted according to the HowNet sentiment dictionary, including positive evaluation words, positive emotional words, negative evaluation words, negative emotional words. Symptom severity was grouped according to the HAMD-17 score: mild depressive symptoms is 8–17 points, moderate depressive symptoms is 18–24 points and severe depressive symptoms is >24 points. Analysis of Variance (ANOVA) was used to analyze the four categories of words among the groups, and partial correlation analysis was used to analyze the correlation between the four categories of word frequencies based on HowNet sentiment dictionary and the HAMD-17 scale to evaluate the total score. Receiver operating characteristic (ROC) curves were used to determine meaningful cut-off values.ResultsThere was a significant difference in negative evaluation words between groups (p = 0.032). After controlling for gender, age and years of education, the HAMD-17 total score was correlated with negative evaluation words (p = 0.009, r = 0.319) and negative emotional words (p = 0.027, r = 0.272), as the severity of depressive symptoms increased, the number of negative evaluation and negative emotional words in clinical interviews increased. Negative evaluation words distinguished patients with mild and moderate-severe depression. The area under the curve is 0.693 (p = 0.006) when the cut-off value is 8.48, the Youden index was 0.41, the sensitivity was 55.2%, and the specificity was 85.4%.ConclusionIn the clinical interview with MDD patients, the number of word frequencies based on HowNet sentiment dictionary may be beneficial in evaluating the severity of depressive symptoms.